You are on page 1of 17

Grammatical Gender and Number Agreement in Spanish:

An ERP Comparison

Horacio Barber1 and Manuel Carreiras2

& The role of grammatical gender and number representa- well as when this happened at the beginning of the sentence
tions in syntactic processes during reading in Spanish was stud- (at the noun), but the last segment of the P600 effect was
ied using the event-related potentials (ERPs) technique. The greater for the middle sentence position, which could indicate
electroencephalogram was recorded with a dense array of differences in the complexity of reanalysis processes. Differ-
128 electrodes while Spanish speakers read word pairs (Experi- ences between grammatical gender and number disagreement
ment 1) or sentences (Experiment 2) in which gender or were found in late measures. In the word pairs experiment, P3
number agreement relationships were manipulated. Disagree- peak latency varied across conditions, being later for gender
ment in word pairs formed by a noun and an adjective (e.g., than for number disagreement. Similarly, in the sentence
faro–alto [lighthouse–high]) produced an N400-type effect, experiment, the last segment of the P600 effect was greater for
while word pairs formed by an article and a noun (e.g., el–piano gender than for number violations. These event-related po-
[the-piano]) showed an additional left anterior negativity tentials (ERPs) effects lend support to the idea that reanalysis
effect (LAN). Agreement violations with the same words in- or repair processes after grammatical disagreement detection
serted in sentences (e.g., El piano estaba viejo y desafinado could involve more steps in the case of gender disagreement,
[the m-s piano m-s was old m-s and off-key]) resulted in a pattern as grammatical gender is a feature of the lexical representation
of LAN–P600. This effect was found both when the violation in contrast to number, which is considered a morphological
occurred in the middle of the sentence (at the adjective), as feature that combines with the stem of the word. &

(Corbett, 1991). Number, on the other hand, is always
Agreement rules have an important role in parsing and considered a conceptual feature signaling the quantity
language comprehension in general, especially in richly of the referent, and therefore the number morphological
inflected languages. They are important and necessary representation of any concept is variable because it can
information for computing grammatical dependencies adopt at least two different forms, singular and plural.
between the different elements of a sentence in order to For this reason, number has been considered a morpho-
build its syntactic structure. Furthermore, agreement logical marker that combines with the stem it modifies
contributes to discourse cohesion, establishing long- (Ritter, 1988), whereas grammatical gender has been con-
distance references across sentences. Among the fea- sidered a fixed property of stem (Harris, 1991). As we
tures used to compute agreement are the marking of will see below, this lexical feature remains as one of the
gender and number, as well as others such as person or most important differences between gender and num-
case. Gender can be a conceptual characteristic or a ber, with relevant consequences at the syntactic level.
formal property of words. Semantic gender is marked Gender and number have been studied from both
for words referring to animate entities, there being a formal and processing perspectives. A concern of the
transparent relationship between the biological sex of formal approach is how to represent gender and num-
the referents and the gender of nouns and pronouns. ber features syntactically. According to some authors
Many languages, in addition to semantic gender, also (e.g., Ritter, 1988), noun phrases can be considered
have gender as an exclusively grammatical property of complements of a higher syntactic projection, the deter-
many nouns. In this case, there is no conceptual basis miner phrase (see Figure 1). Thus, between the noun
for the distinction of gender and this can be considered phrase and the determiner phrase, there could be other
as arbitrarily assigned and invariable because only one functional heads. If a feature is a syntactic head, that
form (feminine or masculine) is assigned to each noun feature is available to the syntactic parser for all syntactic
operations, such as movement, co-referential processes,
or agreement. A feature independently represented in
University of California, 2Universidad de La Laguna, Tenerife, the lexicon must have semantic content and therefore has
Spain to be variable. Number fulfils the requisites of having

D 2005 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 17:1, pp. 137–153
with word pairs. The grammatical priming effect reveals
that a word is recognized faster when preceded by an-
other word that agrees with it than when preceded
by a word that does not agree (Lukatela, Kostic, Todoro-
vic, Carello, & Turvey, 1987). Using this procedure, the
empirical evidence on whether gender and number
agreement mechanisms are similar is mixed. Whereas
two studies have not found differences between gender
and number processing (Colé & Seguı́, 1994; Lukatela
et al, 1987), Faussart, Jakubowitz, and Costes (1999) re-
ported longer recognition times when words disagreed
Figure 1. Formal representation of grammatical gender and number in gender than when they disagreed in number. These
in the determiner phrase structure (see text). authors explain this effect within the framework of
classical models of lexical retrieval, according to which
the target word is retrieved in three successive stages:
variability and semantic content and so can be consid- lexical access, recognition, and integration. The first step
ered as syntactic head. Therefore, some authors, when would be the process by which the right lexical entry is
describing the structure of the determiner phrase, con- located and lexical identification carried out. In the
sider that number is realized as the head of an indepen- second step, the relevant lexical content of that entry
dent functional projection (namely, the number phrase) would be accessed, that is, semantic information, gram-
which is situated between the determiner and the noun matical category, or morphological information. Finally,
(Ritter, 1988). On the other hand, grammatical gender, the third stage would include all postlexical processes of
a formal feature of the noun, is not variable and cannot integration concerning the context. Grammatical agree-
be selected from the lexicon independently of the rest ment would take place at this last stage. Because gender
of the noun features. Consequently, Ritter (1993) argues is a stem inherent feature, integration process failure
that gender can never be the head of its own syntactic regarding gender agreement would make the system go
projection. According to this proposal, under some cir- back to the lexical identification stage in order to check
cumstances there will be different consequences for if the right entry had been selected. Nevertheless, num-
grammatical gender and number syntactic processing. ber is not considered a stem inherent feature, so if num-
There are language production data suggesting that ber agreement is not detected, the processor would only
gender and number agreement are probably processed have to check the final processes of recognition without
or represented in different ways. For example, the rates returning to the initial processes of lexical access. The
of gender and number agreement errors within a lan- difference in access time employed by the processor
guage usually differ considerably, being greater for when detecting a gender as compared to a number in-
number than for gender, which could be interpreted consistency would reflect the additional cost of going
as evidence for different agreement mechanisms (Vig- back one more step.
liocco, Butterworth, & Garrett, 1996). In addition, Igoa, A different proposal that predicts differences for gen-
Garcı́a-Albea, and Sánchez-Casas (1999), in an analysis of der and number agreement during syntactic processing
error production, found that exchange errors affected comes from the studies of De Vincenzi (1999) and De
mostly number suffixes, seldom number and gender Vincenzi and Di Domenico (1999), who examined the
simultaneously and never gender alone. These differ- use of morphological information in pronoun resolution
ences in the error rates suggest that gender is retrieved mechanisms in Italian. They found that although both
directly from the lexicon and assigned to the phrase number and gender are used to activate an antecedent
structure together with the lemma, because grammatical at the end of a sentence, only number information re-
gender is part of the lemma, while number is derived activates the antecedent immediately after pronouns
by rule. Therefore, gender morphemes should be within the sentence. They explain these differences in
stranded less often than number morphemes, which number and gender information relying on serial models
are hypothesized to be assigned directly to the phrases of language processing (Frazier, 1987), according to
as closed class items (Bock & Levelt, 1994; Garrett, which the construction of structural representations
1982). This proposal has been recently supported by initially employs only information relevant to syntactic
another production study in Spanish in which the analysis. In this first step, only features with autono-
number of combined gender and number errors was mous projections on syntax, such as number (following
not greater than that expected at random, showing that Ritter, 1993), would be used to establish coherence be-
these features are processed independently (Antón- tween pronoun and noun. On the contrary, gender in-
Méndez, Nicol, & Garret, 2002). formation, which does not have separate identity inside
Processing of gender and number agreement in com- the syntactic structure, would only be considered at a
prehension has been studied using grammatical priming later stage with lexical and semantic information.

138 Journal of Cognitive Neuroscience Volume 17, Number 1

In sum, the previous studies seem to suggest that the P600 effect was preceded by a LAN effect (Gunter,
there are differences between gender and number in Friederici, & Schriefers, 2000). This LAN–P600 pattern
their representation at the lexical level. However, it is was confirmed in Spanish when agreement violations of
not yet clear how and when these representational differ- two types of gender (semantic and grammatical) were
ences influence agreement during syntactic processing. compared in the same experiment. Although additional
differences between the semantic and the grammatical
gender processing were found, the LAN–P600 response
Electrophysiological Studies
to the agreement violations was equal to both types of
Agreement processes have also been studied with the gender (Barber, Salillas, & Carreiras, 2004).
ERP technique, as its multidimensional character and Münte and Heinze (1994) undertook a series of ex-
excellent temporal resolution make this technique es- periments in which the effect of grammatical priming in
pecially suitable for the study of when agreement rela- word pairs was analyzed. Among other types of syntactic
tionships are made during reading. Pioneer studies have relationships, gender agreement was manipulated in
shown different effects of grammatical agreement viola- pairs formed by an article followed by a noun. The mor-
tions. Kutas and Hillyard (1983) found an increase of phosyntactic effects were compared with other phono-
negativity between 200 and 500 msec in anterior zones logical and semantic effects. Moreover, the experiments
as a response to the reading of number agreement er- were undertaken in different languages (German and
rors between subject and verb in a sentence. In contrast, Finnish) and with different tasks (grammatical judgment
Hagoort, Brown, and Groothusen (1993) described a or lexical decision). Results showed that both agreement
P600/SPS effect (an increase of positivity from 500 msec) and verb inflection errors produced anterior distributed
in response to the same type of violation. Similar ef- negativities and that these effects seem to reflect rela-
fects to those described by Kutas and Hillyard have been tively automatic processes that are independent of the
linked to different aspects of syntactic analysis and la- task. The authors point to the resemblance of these
beled left anterior negativity (LAN) (Friederici, 1995). negativities with the LAN effect found in sentence con-
This anterior negativity could be reflecting, in fact, some texts. Inversely, the effects produced by phonological or
kind of verbal working memory load, as proposed by semantic relationships produced negativities with more
some authors (Kluender & Kutas, 1993). The P600/SPS posterior distributions. In a similar study, Barber and
effect has been also linked to syntactic parsing and to Carreiras (2003) found an effect of the N400 type in re-
reanalysis processes during syntactic processing. Var- sponse to gender or number disagreement in adjective–
ious authors have differentiated at least two phases noun word pairs. No differences between gender and
in this effect which could be related to functionally dif- number effects were observed in this time window.
ferent states of the parser: the first one, between 500 However, the P3 component that followed this effect
and 750 msec, shows a broad distribution along the presented a longer latency in the case of gender dis-
midline of the scalp, and the second one, after 750 msec, agreement. The sensitiveness of the amplitude of the P3
is more localized in parietal areas (Hagoort, Brown, & towards characteristics such as the probability of stim-
Osterhout, 1999). ulus occurrence or its subjective relevance for the task
In another study, Osterhout and Mobley (1995) com- indicates that when this component is produced, the
pared number agreement violations between subject necessary processes for the categorization of the stimu-
and verb, between a reflexive pronoun and its anteced- lus must have finalized (Donchin, 1979). Although the
ent, and between a personal pronoun and its ante- classic effect of this component has been located around
cedent. In the three cases, the words that infringed 300 msec, several studies have shown that this latency
agreement rules produced an amplitude increase at can be retarded depending on the complexity of the
the P600 time window. Only in the case of noun–verb stimulus (Kutas, McCarthy, & Donchin, 1977). For this
agreement was an additional LAN-type effect found. In reason, this component usually shows up at around
these experiments, pronouns were also manipulated to 500-msec delay with linguistic stimuli. These same la-
produce gender agreement violations. The effects of tency variations can occur even with the same stimuli
gender violations were never different from those of when its categorization difficulty varies. For example,
number (i.e., P600 effect), and both were different from in lexical decision tasks, pseudowords produce higher
the semantic violations. Semantic violations usually pro- latencies than words do (Bentin, McCarthy, & Wood,
duce a significant enhancement of the N400 component, 1985), and repeated words have shorter latencies than
a negative waveform associated to lexical–semantic inte- those not repeated (Rugg, 1985). However, it should
gration processes (see review in Kutas & Federmeier, be borne in mind that the relationship between the
2000). Other studies have also reported the P600 effect P600 and the P3 is an open issue. Some authors have
in response to grammatical gender agreement violations claimed that the P600/SPS belongs to the family of P300
in languages in which (in contrast with English) gender effects (Coulson, King, & Kutas, 1998) although this
is not exclusively semantically based (e.g., Hagoort & interpretation is not shared by others (Osterhout &
Brown, 1999, in Dutch). In addition, in a German study, Hagoort, 1999). The latency differences in the experiment

Barber and Carreiras 139

of Barber and Carreiras (2003) could be related, accord- ERP components associated to syntactic processing, as
ing to the authors, to reanalysis or rechecking processes well as the influence of the amount of syntactic infor-
occurring after morphological integration failure. Thus, mation available during the course of sentence reading.
reanalysis processes after gender disagreement detec-
tion could be associated with a higher cost because, as
suggested by Faussart et al. (1999), gender is a lexical
feature and therefore checking of gender agreement in-
cludes both integration and lexical access processes, Grammatical priming was analyzed in word pairs in
while number disagreement only requires checking of which gender and number agreement between an article
the integration step. and a noun (e.g., el–piano [the–piano]), and between a
noun and an adjective (e.g., faro–alto [lighthouse–
high]) were manipulated. This setting provides an op-
The Present Study
portunity to investigate gender and number in the same
In summary, the behavioral data as well as the ERP data stimuli and with the simplest syntactic context. In a
on gender and number agreement provide a mixed pic- previous experiment, word pairs composed by a noun
ture. The behavioral data are contradictory but they in- and an adjective showed the same posterior negativity
dicate that there could be differences between gender (N400) in response to gender and number disagreement
and number, at least in their representation. The ERP (Barber & Carreiras, 2003). In this new experiment, we
data from different experiments seem to suggest that expected that the article–noun pairs (which, in contrast
gender and number disagreement produce, at least, with adjective–noun pairs, can be considered a noun
qualitatively similar effects. Nonetheless, the ERP data phrase) would trigger more syntactic processing and
available at present come from different designs in dif- therefore a more anterior effect, similar to that found
ferent languages. As far as we know, only the study of by Münte and Heinze (1994) with this type of stimuli.
Osterhout and Mobley (1995) has directly compared the This way, we checked possible differences between
ERP effects of gender and number agreement violations gender and number in this very early agreement pro-
in the same experiment. Because this experiment was cessing, as should be expected following the proposal of
performed in English, only semantic gender could be De Vincenzi (1999) and De Vincenzi and Di Domenico
manipulated. However, Spanish is a language with a rich (1999). In addition, we expected to replicate previously
morphology in which agreement relationships in gen- reported late effects (i.e., P3 latency variations) that
eral, and grammatical gender agreement in particular, support the idea of different reanalysis costs due to
are highly relevant in order to build the correct syntactic differences in the lexical representation of gender and
structure of sentences. This characteristic of the Spanish number (Barber & Carreiras, 2003).
language leads us to the present research, in which the
processing of both types of agreement was compared
under the same circumstances with Spanish speakers.
The main goal of the present experiments was to study Figure 2 shows grand averages of the agreement, the
the possible differences in the processing of gender and gender disagreement, and the number disagreement
number, and to discriminate between the effects due conditions for the two types of word pairs. For a clearer
to different lexical representations that could trigger dif- description of the data, in these and other figures, only
ferent reanalysis processes (Faussart et al., 1999) and one representative electrode of each analysis region
effects due to time-course differences in agreement pro- is plotted. Visual inspection reveals clear differences
cessing during the building of a syntactic structure (De in the responses to the agreement conditions with re-
Vincenzi, 1999; De Vincenzi & Di Domenico, 1999). We spect to the disagreement conditions. Targets for the
compared the ERP effects of gender and number agree- disagreement word pairs start to differ from those of
ment violations in the context of lexical integration and the well-formed pairs at 250–300 msec. Between 300 and
syntactic processing. In the first experiment, we com- 500 msec, disagreement waves were more negative-
pared de effect of agreement in article–noun word pairs going than those to the agreement condition, with the
(e.g., el–piano [the–piano]) and in noun–adjective word largest amplitude differences at 400 msec. Furthermore,
pairs (e.g., faro–alto [lighthouse–high]), while in the in this time window, no differences between disagree-
second experiment the agreement between the same ment conditions can be appreciated. As can be seen
words was manipulated at the beginning of sentences in Figure 3, the disagreement effects are broadly distrib-
(e.g., El piano estaba viejo y desafinado [the m-s piano uted over the scalp. The effects associated to article–
m-s was old and off-key]), or in the middle of sentences noun word pairs are localized both in the frontal areas
(e.g., El faro es alto y luminoso [the m-s lighthouse m-s and in the central and posterior electrodes, while noun–
is high m-s and bright]). This way, comparing the results adjective pair effects are more restricted to central
of both experiments, we could also trace the impact of and posterior areas. The subtraction of the respective
the sentence context and reading expectations in the topographical maps shows that differences across the

140 Journal of Cognitive Neuroscience Volume 17, Number 1

Figure 2. Grand averages of the target words for the agreement, gender disagreement, and number disagreement conditions, in article–noun pairs
(left) and in noun–adjective pairs (right). Vertical line marks the onset of target word presentation (t = 0). Six representative electrodes are
represented (see locations in Figure 7).

two types of word pairs are localized in the left ante- nent peaks earlier in the agreement condition than in
rior area. both disagreement conditions. More detailed inspection
The N400 component was followed by the P3 com- of averaged electrode groups revealed longer peak
ponent, which was present in all conditions and showed latency for the gender disagreement condition as com-
similar amplitudes across these. However, systematic pared with the number disagreement condition and sta-
peak latency variations were appreciated. This compo- tistical analysis confirmed these observations.

Figure 3. Topographical
maps obtained by
interpolation from
128 electrodes at 350, 400, and
450 msec. Maps were
computed from values
resulting from the subtraction
between agreement and
disagreement waves (gender
and number conditions
collapsed). The first row shows
the effects in article–noun
word pairs, the second row in
noun–adjective pairs, and
the third row shows the
differences between the effects
of both types of word pairs.

Barber and Carreiras 141

300–500 msec Time Window: Amplitude Differences and number disagreement) and the ‘‘electrode region’’
factor (left and right). Because of the posterior distribu-
Mean amplitude values corresponding to the 300–
tion of the P3 component, only the posterior electrode
500 msec time epoch were introduced in an ANOVA
regions were included. This analysis yielded a significant
with the ‘‘grammatical agreement’’ factor (agreement,
main effect of grammatical agreement [F(2,18) = 34.74;
gender disagreement, and number disagreement), the
p < .001]. Post hoc tests showed significant differences
‘‘pair type’’ factor (article–noun and noun–adjective), the
in the article–noun pairs between the agreement con-
‘‘electrode region’’ factor (anterior, central, and poste-
dition (543 msec) as compared to both the gender
rior), and the ‘‘hemisphere factor’’ (left and right). This
disagreement condition [602 msec; F(1,19) = 22.58; p <
analysis revealed a marginally significant four-way inter-
.001] and the number disagreement condition [582 msec;
action [F(4,18) = 2.41; p = .08; e = 0.64], which was
F(1,19) = 13.87; p < .01], and between the gender
confirmed after the data normalization [F(4,18) =2.75;
disagreement and number disagreement conditions
p = .05; e = 0.64]. Post hoc tests showed reliable differ-
[F(1,19) = 7.56; p < .05]. Similar effects were found in
ences between the agreement condition and the gender
the comparisons with noun–adjective pairs, between
disagreement condition [F(1,21) = 33.24; p < .001] and
agreement (533 msec) and gender disagreement
between the agreement condition and the number dis-
[598 msec; F(1,19) = 51.79; p < .001], agreement and
agreement condition [F(1,21) = 30.55; p < .001] in the
number disagreement [576 msec; F(1,19) = 19.33; p <
article–noun pairs, as well as similar differences between
.001], and between gender disagreement and number
the agreement condition and the gender disagreement
disagreement conditions [F(1,19) = 9.12; p < .01].
condition [F(1,21) = 9.13; p < .01] and between the
agreement condition and the number disagreement con-
dition [F(1,21) = 7.64; p < .01] in the noun–adjective
pairs. However, no differences between the disagree-
ment conditions were found in any case (F < 1). Table 1 The results show that grammatical disagreement for
shows the F values and significance levels of gender each word pair produced a remarkable effect between
and number disagreement effects in each type of word 300 and 500 msec. Disagreement conditions presented
pairs and each electrode region. Table 1 indicates differ- waves with higher negative amplitude than those of
ent scalp distributions of the effects across both types of agreement. These differences were distributed over
word pairs, while article–noun pairs show reliable effects central and posterior areas in both types of word pairs,
in the left anterior electrode region that do not happen while left anterior effects are found only for the article–
with the noun–adjective pairs. noun pairs.
The differences that were found in central–posterior
areas fit with the classical N400 effect. This effect has
P3 Component: Peak Latencies Analyses
been reported associated to semantic priming between
Maximum amplitude latencies between 450 and 750 msec words and has usually been associated with semantic
were introduced in an ANOVA with the ‘‘grammatical integration processes (see Osterhout & Holcomb, 1995,
agreement’’ factor (agreement, gender disagreement, for a review). However, the manipulation of other word

Table 1. F Values and Significance Levels Resulting from Post Hoc Comparisons between the Average Amplitude Values of the
Experimental Conditions (Agreement versus Gender Disagreement and versus Number Disagreement), for Each Word Pair Type
(Article Plus Noun and Noun Plus Adjective) and for Each Electrode Region (Anterior, Central, and Posterior) in Each Hemisphere
(Left and Right)
Article–Noun Word Pairs Noun–Adjective Word Pairs

Agreement vs. Agreement vs. Agreement vs. Agreement vs.

Gender Disagreement Number Disagreement Gender Disagreement Number Disagreement

df = 1, 19 Left H. Right H. Left H. Right H. Left H. Right H. Left H. Right H.

Anterior 30.04*** 15.43** 28.54*** 11.55** –a 7.22* –a 7.05*

Central 37.00*** 25.30*** 27.46*** 15.42** 3.60 15.80** 5.06* 18.33**
Posterior 15.21** 10.33** 15.68** 13.06** 5.86* 14.05** 3.70 8.18**

***p > .001.

**p > .01.
*p > .05.
F < 1.

142 Journal of Cognitive Neuroscience Volume 17, Number 1

features can result in a similar N400 effect. For instance, pairs could be the result of a syntactic integration
rhyme matching between words can result in a reduc- attempt that does not seem to happen in the case of
tion of the N400 component, but these phonological noun–adjective pairs.
effects, in contrast with semantic ones, have been found
to be very dependent on the task that the subject has to
Gender versus Number
perform, and they are only found when subjects are
asked to carry out some kind of phonological judgment The comparison between the gender and number dis-
task (Radeau, Besson, Fonteneau, & Luis-Castro, 1998; agreement effects did not reveal any reliable difference
Rugg & Barrett, 1987). Therefore, the processes associ- either in the size of the effects or in their distribution.
ated with the N400 component should be understood as However, this lack of difference could be due to the lack
implying the integration of different types of features of a real syntactic context or due to the word pairs not
(e.g., lexical, conceptual, etc.) when this is requested. having enough syntactic complexity to trigger adequate
In the present experiment, the task that the subjects processing. In order to explore these two possibilities,
had to do (agreement judgment) could be accomplished Experiment 2, in which the same words were manipu-
by comparing the morphological features of the two lated in a sentence context, was planned.
words, so the N400 modulation could be related to the In spite of the fact that no amplitude differences
integration of these specific lexical features. However, between gender and number effects were found, anal-
it is worth noting that this lexical integration is nei- ysis of the latency variations of the P3 peaks replicated
ther semantic in nature nor necessarily implies syntactic previous results with noun–adjective word pairs (Barber
parsing, as the building of a syntactic structure is not & Carreiras, 2003). That is, disagreement conditions
necessary. showed longer latencies than the agreement condition.
In addition, in the article–noun word pairs, the de- In addition, although the graphics of single electrodes
scribed disagreement effect is not restricted to posterior do not show clear differences between the peak laten-
areas, but covers anterior areas as well. The different cies of both disagreement conditions, statistical analysis
types of vocabulary classes involved in the integration of the mean values of groups of electrodes revealed that
of both kinds of word pairs could trigger the activation the gender disagreement produced longer latencies
of different types of neural networks distributed across than the number disagreement condition. N400 and
different brain areas. The representations of content P300 components usually present similar latencies so it
words (nouns and adjectives) could be stored in poste- is important to consider the possible effect of some
rior areas while function word representations would be partial component overlapping. This way, P300 latency
more restricted to left anterior areas (Pulvermüller, variations could be the result of previous effects. How-
1999). Thus, the integration of a function word plus a ever, no latency differences across conditions can be
content word would result in the activation of both appreciated at the onsets or peaks of the previous
anterior and posterior areas. However, the presence of negativities. Generally speaking, a P3 appears each time
function words could be inseparable from the triggering the task requires a binary-type decision (Donchin &
of syntactic processing. In this way, the broader distri- Coles, 1988). Several studies have shown that P3 latency
bution of the effect associated to article–noun pairs can be retarded depending on the complexity of the
could be the result of the overlapping of two differ- stimulus or the categorization difficulty that the subject
ent effects; the previously described N400 effect with must confront (Bentin et al, 1985; Rugg, 1985; Kutas,
central–posterior distribution and another one over the McCarthy, et al, 1977). Therefore, changes in the laten-
left anterior area. The anterior effect could be related to cies of the P3 component in our data could be reflecting
the anterior negativities that have been associated with differences in the categorization of word pairs to pre-
syntactic processes, such as those reported by Münte pare the response. In the case of grammatically agreeing
and Heinze (1994) in response to grammatical violations word pairs, decision can be taken immediately after
between article–noun word pairs. This effect would integration. However, for pairs in which such integration
reflect the failure in the integration of the two words is impossible, the decision has to be postponed until all
in order to build a syntactic unit. It is worthy of note possible checking and integration attempts have been
that, in Spanish, noun–adjective word pairs (because of made. This way, differences between gender and num-
the lack of the determiner) cannot constitute an auton- ber disagreement could be the result of the different
omous syntactic unit. In contrast, article–noun word costs of these rechecking processes. In relation to the
pairs can be considered as a noun phrase, and can con- syntactic analysis of sentences, different authors have
stitute an adequate stimulus that triggers syntactic inte- proposed that reanalysis and repair processes are re-
gration processes. In addition, the article–noun pair can flected in the P600 component (e.g., Friederici, 1995). If
and usually does constitute the beginning of a new gender disagreement produces costlier reanalysis than
sentence, so it is from the very first moment an adequate number disagreement, greater amplitude of the P600
entity to open a new syntactic tree. Therefore, the component in response to gender agreement violations
presence of an effect in anterior zones in article–noun should be expected with respect to that produced by

Barber and Carreiras 143

number agreement violation. This hypothesis was ex- are represented in different ways during syntactic pro-
plored in the following experiment. cessing (De Vincenzi & Di Domenico, 1999), a larger
effect of number violation was expected than that of
gender violation in the initial syntactic parsing, producing
a larger effect over the LAN. However, in Experiment 1,
no differences were found in the anterior negativity.
In this experiment, the effects of agreement violations Therefore, in this new experiment, it was possible to
were analyzed in more complex syntactic structures and check if the lack of these differences was due to the
in a more natural reading process. The same words as in lack of a real syntax context. In addition, considering the
Experiment 1 were presented opening a sentence (e.g., variations in latency of the P3 component found in Ex-
el piano estaba viejo y desafinado [the m-s piano m-s periment 1, and that we have related these changes to
was old and off-key]). This way, it was possible to rechecking processes, a larger amplitude of the P600
compare the effect of the isolated word pairs of Exper- associated to gender violations could be hypothesized
iment 1 (i.e., el–piano [the–piano]) with these pairs in comparison with number violations, due to the re-
localized at the beginning of the sentence, in order to analysis processes being costlier when the processor has
analyze the effect of the sentence context in reading. to deal with lexical features (Faussart et al, 1999). Fi-
Additional sentences, made from the noun–adjective nally, it was expected that the effect of reanalysis would
word pairs (e.g., El faro es alto y luminoso [the m-s be greater when the violations were localized across
lighthouse m-s is high m-s and bright]), presented the phrases, in the middle of the sentence, than when they
same type of manipulations localized after the verb were at the beginning of the sentence, within the same
(violating the noun phrase and a predicative adjective phrase.
agreement), producing the violations across phrases in
the middle of the sentence. Therefore, the impact of
gender and number agreement violations in two differ-
ent moments of sentence processing was compared. Figure 4 shows the grand averaged waveforms cor-
Under these circumstances, the P600 effect was expected responding to the agreement manipulations at the be-
in response to all agreement violations (Osterhout & ginning and at the middle of the sentences separately.
Mobley, 1995) and a LAN effect, at least, for the gender Visual inspection of both figures reveals clear differ-
agreement violations (Barber, Sallilas, et al., 2004). Fol- ences in the responses to disagreement conditions with
lowing the models that propose that gender and number respect to agreement conditions. Between 300 and

Figure 4. Grand averages of the target words for the agreement, gender disagreement, and number disagreement conditions, at the beginning
(left) and at the middle of the sentence (right).

144 Journal of Cognitive Neuroscience Volume 17, Number 1

450 msec, disagreement waves were more negative- and 500–700 and 700–900 msec for the P600 effect.
going than those in the agreement conditions. This Omnibus ANOVAs were performed, including the ‘‘gram-
effect is similar for both gender and number violations matical agreement’’ variable as a within-factor with three
and is present at anterior electrodes, especially on the levels (agreement, gender disagreement, and number
left side of the scalp, fitting with the previously de- disagreement), the ‘‘sentence position’’ factor with two
scribed LAN effect. In both figures, the negative effect levels (beginning and middle), the ‘‘electrode region’’
is followed by a typical P600 effect, showing larger am- factor with three levels (anterior, central, and posterior),
plitudes for the disagreement conditions than for the and the ‘‘hemisphere’’ factor with two levels (left and
agreement ones. The P600 effect starts at 500 msec and right).
is maintained through more than 400 msec. Between
500 and 700 msec, the P600 effect was distributed along
300 – 450 msec Time Window: LAN
anterior and posterior areas. In a second stage, between
700 and 900 msec, the P600 effect is only localized over In this time window, a three-way interaction was found
posterior areas and lateralized to the right side of the involving the factors ‘‘grammatical agreement,’’ ‘‘elec-
scalp. These changes in the topography of the P600 trode region,’’ and ‘‘hemisphere’’ [F(4,20) = 6.03; p <
effect across time fit with the two phases previously .01; e = 0.55] and this was confirmed after data normal-
proposed by Hagoort, Brown, and Osterhout (1999), so ization [F(4,20) = 4.3; p < .05; e = 0.63]. Post hoc tests
we will consider these two phases separately in the fol- showed significant differences between the agreement
lowing analysis. The topographical distribution of the condition and the gender disagreement condition just in
different effects is represented in Figure 5 collapsing the left anterior [F(1,23) = 10.15; p < .01] and the left
the gender and number violations across the two posi- central [F(1,23) = 10.37; p < .01] regions. In a similar
tions of the sentence. The comparison between the way, differences between the agreement condition and
gender and number disagreement conditions reveals the number disagreement condition were only present
differences only in the second phase of the P600 effect. in the left anterior region [F(1,23) = 5.58; p < .05].
In this last time window, the effect of gender viola- Contrasts between both disagreement conditions re-
tion seems to be greater as compared with number vealed no significant differences.
violation. As to the sentence position of the agreement
violations, the LAN effect is of equivalent size in both
500–700 msec Time Window: P600a
positions of the sentence, while the last segment of
the P600 effect is larger in the violations located in the The ANOVA revealed a three-way interaction between the
middle of the sentence and especially at the posterior factors ‘‘grammatical agreement,’’ ‘‘electrode region,’’
electrodes. Comparison of the disagreement effects and ‘‘hemisphere’’ [F(4,20) = 5.8; p < .001; e = 0.86],
(subtraction between the agreement and disagreement interaction that was maintained after the data normaliza-
waves) across the two sentence positions can be found tion [F(4,20) = 4; p < .01; e = 0.83]. However, the
in Figure 6. ‘‘grammatical agreement’’ variable was not modulated
Analyses were carried out in three different tempo- by the ‘‘sentence position’’ factor (F < 1). F values and
ral windows on the basis of calculations of mean ampli- significance levels of the post hoc tests (comparisons
tudes; between 300 and 450 msec for the LAN effect, between the different experimental conditions in each

Figure 5. Topographical maps

obtained by interpolation at
400, 600, and 800 msec
corresponding to LAN, P600
first phase, and P600 second
phase. Maps were computed
from values resulting from the
subtraction between the gender
disagreement and agreement
condition (top row) and
number disagreement less
agreement condition
(bottom row).

Barber and Carreiras 145

Figure 6. Difference waves (agreement less disagreement) comparing gender (left) and number (right) across both sentence positions.

electrode region) are listed in Table 2. It can be seen that cal agreement,’’ ‘‘electrode region,’’ and ‘‘hemisphere’’
comparisons between the disagreement conditions did [F(4,20) = 8.44; p < .01; e = 0.62; with normalized data
not show any reliable differences in this time window. F(4,20) = 8.14; p < .01; e = 0.62]. F values and
significance levels of the post hoc tests (comparisons
between the different experimental conditions in each
700–900 msec Time Window: P600b
electrode region) are listed in Table 2. This table shows
As in the previous window, an ANOVA showed the same how the disagreement effects are located in central and
three-way interaction between the factors ‘‘grammati- posterior regions and are greater in the right hemi-

Table 2. F Values and Significance Levels Resulting from Post Hoc Comparisons between the Average Amplitude Values of the
Experimental Conditions (Agreement versus Gender Disagreement and versus Number Disagreement, and Gender Disagreement
versus Number Disagreement), for Each Time Window of the P600 Effect (from 500 to 700 msec and from 700 to 900 msec) and for
Each Electrode Region (Anterior, Central, and Posterior)
Agreement vs. Agreement vs. Gender Disagreement vs.
Gender Disagreement Number Disagreement Number Disagreement
df = 1, 23 Left H. Right H. Left H. Right H. Left H. Right H.

500 –700 msec Anterior –a 3.62 –a 5.74* –a –a

Central 14.61* 22.66*** 17.76*** 16.1*** –a –a
Posterior 62.87*** 51.53*** 39.27*** 26.1*** 1.25 2.35
# a # a a
700 –900 msec Anterior 5.85 – 7.33 – – –a
Central 20.29*** 64.90*** 17.00*** 35.82*** 1.42 5.84*
Posterior 106.9*** 137.0*** 65.28*** 70.27*** 4.78* 6.05*

***p > .001.

*p > .05.
p > .05 (opposite direction).
F < 1.

146 Journal of Cognitive Neuroscience Volume 17, Number 1

sphere. Even though there are significant differences in agreement, whereas the article–noun disagreement at
the left anterior regions, mean values indicate that they the beginning of the sentence was a within-phrase
have the opposite direction to the posterior ones, so disagreement. Thus, P600 differences across positions
they seem to reflect the polarity inversion of the right could be reflecting the fact that mid-sentence violations
posterior effect in the left anterior areas. Comparisons of involve more cost in reanalysis processes after the
both disagreement conditions showed significant differ- violation detection. The first P600 phase could reflect
ences in the posterior regions of both hemispheres as the detection of the violation or the failure in syntactic
well as in the central region of the right side, indicating integration, these processes being similar in both posi-
that amplitudes were larger for gender than for num- tions. On the other hand, the second segment of the
ber. In addition to these effects, there was a significant P600 would reflect the reanalysis processes which would
three-way interaction between the factors ‘‘grammatical be costlier in the case of the mid-sentence position
agreement,’’ ‘‘sentence position,’’ and ‘‘electrode region’’ because of the greater complexity of the agreement.
[F(4,20) = 8.55; p < .01; e = 0.53], an effect that However, we cannot dismiss an alternative interpreta-
also remained after the data normalization [F(4,20) = tion in terms of vocabulary type, due to the fact that the
4.11; p < .05; e = 0.53], indicating the greater effect target words are nouns in the beginning position and
over posterior regions of the violations located in the adjectives in the middle position. Although the question
middle of the sentence as compared with those that of the exact source of these differences is not central to
happened at the beginning of the sentence. the main goal of the present study, it is relevant by itself
and so should be addressed in future experiments.
Gender versus Number
Agreement violations produced two effects that have
been typically associated with syntactic processing; a Comparison between gender and number effects show-
negativity between 300 and 450 msec with a left anterior ed differences in the second phase of the P600 effect.
distribution that fits with the so-called LAN effect, and an These differences are located in the posterior areas and
inverse polarity effect that begins around 500 msec and are larger on the right side of the scalp. In contrast, the
stays for more than 400 msec, which corresponds with LAN effect and the first phase of the P600 effect were
the P600 effect. The P600 effect was greater over the similar in latency, distribution, and amplitude for gender
right side of the scalp and over posterior areas, espe- and number violations. These data point to very late
cially after 700 msec. differences in the detection of gender and number
According to previous interpretations, the LAN effect disagreement during syntactic processing. Gender and
could be reflecting the detection of a mismatch between number differences were found in the same time win-
morphosyntactic features, the difficulty of integrating dow as the effects of sentence position. If we assume
these characteristics in a syntactic structure (Münte, that this late window is sensitive to the costs associated
Matzke, & Johannes, 1997; Friederici, 1995), or an with reanalysis processes, then gender violations would
increase of the working memory demands implied in trigger costlier reanalysis processes than number agree-
these processes (Kluender & Kutas, 1993). In a similar ment violations. This idea fits with the late differences
way, the P600 could also reflect the impossibility of a found in Experiment 1 and is considered in more depth
later integration process, different costs of reanalysis and in the general discussion.
repair processes (Hagoort, Brown, & Osterhout, 1999;
Friederici, 1995) or a more general activation associated
to anomaly detection (Coulson et al., 1998). In addition,
some authors have proposed that different effects (with In two different experiments, the effects of gender and
different generators) could contribute in the early and number disagreement in visually presented words were
late phases of the P600 effect, the first phase being analyzed with the ERP technique. Disagreement in arti-
associated with diagnosis or syntactic integration diffi- cle–noun word pairs resulted in a broadly distributed
culty and the second one with repair or reanalysis negativity that could be the result of the overlapping
processes (Friederici, Mecklinger, Spencer, Steinhauer, of an N400 effect and an anterior negativity, and the
& Donchin, 2001; Hagoort & Brown, 2000). agreement violations during sentence reading resulted
As to the position in which the agreement violation in a LAN–P600 pattern. Changes in the topographic
was located, while the LAN effect showed similar mag- distribution of the P600 and the different sensitivity to
nitude across positions, the second phase of the P600 our manipulations (type of agreement violation and
effect was greater for the violations located in the middle sentence position of the violation) at different times
of the sentence. The violations in the middle of the support the division of this effect into two different
sentence involved a disagreement between a predicative subeffects probably associated to different generators.
adjective and a noun phrase with the verb between In Experiment 1, the N400 effect reflects the attempt
these and so can be considered as across-phrases dis- at integration of the lexical features of the words.

Barber and Carreiras 147

Because the N400 effect is not present when subjects word pair experiment because the task can be per-
read the same violations in a sentence context (Exper- formed matching the lexical features of the words.
iment 2), this lexical integration could be related to the
word matching that participants have to carry out in
order to perform the task, rather than with real syntactic Gender versus Number
integration. In addition to the central–posterior effects,
article–noun word pairs also produced disagreement As we described above, whereas grammatical gender is
effects in the left anterior area, so we have proposed usually considered a lexical feature of the representation
the overlapping of an anterior negativity in this type of of some words, number is considered an autonomous
word pairs. This anterior negativity and the LAN effect of feature that combines with the stem of the word. For
Experiment 2 could be reflecting related underlying this reason, it has been proposed that only number
processes. In both experiments, anterior effects can be features have autonomous representations during syn-
associated with failure in the initial syntactic parsing or tactic analysis and therefore only number could be
even with some kind of additional working memory load considered in early syntactic processing. On the other
associated with this failure. This attempt at syntactic hand, the different lexical representations of gender and
integration would be automatically triggered by any number features could result in different reanalysis
adequate syntactic stimuli, even when the construction processes after the detection of agreement violations.
of a syntactic structure is unnecessary, as in the case of Models that propose different roles of gender and
the article–noun word pairs of Experiment 1. number features during the early stages of syntactic
The P600 effect was present only when participants processing (De Vincenzi, 1999; De Vincenzi & Di Dome-
read sentences but not when they read isolated word nico, 1999) predict different effects of gender and
pairs, where only P3 latency differences were found. number disagreement, especially in relation to anterior
These differences between the experiments could be negativities, as these have been related with first pass
due to the fact that, in the sentence experiment, readers syntactic processing (Friederici, 1995). Thus, gender
can categorize a sentence as wrong immediately after disagreement might produce weaker negative effects
the reading of a violation (which would result in a P3), or none at all. The results of these two experiments
but in the case that they do not find a violation, they cannot support this hypothesis, as no differences were
have to delay their decision (with no P3 at this point) found between the negative effects related to the two
until the end of the sentence. However, other studies types of disagreement, either in the word pairs ex-
have found similar P600 effects when the agreement periment or in the sentence experiment. In both ex-
violation (and therefore the task resolution) was always periments, the early effects of gender and number
located at the end of the sentence (Hagoort, Brown, & agreement violations produced effects with the same
Groothusen, 1993) or even when no task at all was amplitude, latency, and distribution, indicating that the
requested (Osterhout & Mobley, 1995). In this respect, detection of such violations could involve at least some
even when the P600 effect is only present in the common process. Obviously, equivalent effects on the
sentence experiment, our data cannot resolve the ques- surface of the scalp could be a result of changes orig-
tion of whether the P600 and P3 are manifestations of inating in different neural generators and additional
similar underlying processes or not. An alternative inter- processes were possibly not detected by our technique.
pretation can be made attending to the different com- However, the existence of early effects derived from
plexity of the syntactic analysis across experiments. It is gender disagreement indicates that gender information
worth noting that in the case of the target nouns, these is used in the first stages of syntactic processing.
provided exactly the same morphosyntactic (and seman- The P600 has also been considered a manifestation of
tic) information when presented in a word pair (Exper- syntactic integration difficulty (Hagoort & Brown, 2000;
iment 1) as when they were presented at the beginning Kaan, Harris, Gibson, & Holcomb, 2000). However, in
of a sentence (Experiment 2). Differences in the P3 our data, gender and number differences were found at
latency have been explained in terms of the categoriza- a very late stage of this effect, while the early phase was
tion that subjects have to do just after the reading of the similar in both gender and number violations. Although
second word. On the other hand, the P600 effect could differences between gender and number are predicted
be related with a more controlled process that depends in initial syntactic processing, our effects after 700 msec
on reader expectations. Thus, when the participants seem to be too late to be reflecting this initial syntactic
know that they are reading a sentence, they start to stage.
build or predict a syntactic structure right from the In both experiments, differences in the effects associ-
beginning of the sentence. This syntactic structure ated to gender and number violations are found at late
would be used not only to understand the sentence time windows. In the word pairs experiment, P3 latency
but also to perform the task, especially when they find is longer to gender as compared to number disagree-
an agreement violation in the middle of the sentence. ment, indicating that the response-related decision (the
The syntactic structure would not be necessary in the categorization of a word pair as wrong) takes place later

148 Journal of Cognitive Neuroscience Volume 17, Number 1

for gender disagreement. In a similar way, in the sen- 29 years (mean = 22.7 years) in Experiment 2. All par-
tence experiment, the size of the effect of gender agree- ticipants were right-handed, as assessed by an abridged
ment violation is greater than the number effect after Spanish version of the Edinburgh Handedness Inven-
700 msec, indicating a greater impact of gender viola- tory (Oldfield, 1971): LQ > + 50. Seven of the partici-
tion on the later syntactic processes. Following the pro- pants in Experiment 1 and six in Experiment 2 reported
posal by Faussart et al. (1999), when an inconsistency is having left-handed relatives. Subjects were volunteers
detected in agreement relations, in the case of gender and received course credit for their participation. None
failure, the processor would have to check both syn- of the subjects of Experiment 2 had participated in Ex-
tactic integration processes and lexical access (because periment 1.
gender is a lexical feature). In contrast, in the case of
number information posing a problem, the processor
EEG Recording and Analysis
would only have to check the syntactic integration pro-
cesses. Therefore, differences in the reanalysis processes Scalp voltages were collected from Ag/AgCl electrodes
would explain the higher latencies in the case of gender using a 128-chanel Geodesic Sensor Net (see Figure 7).
disagreement as compared to number disagreement in The vertex electrode was used as reference (REF), and
the word pairs experiment, as well as the greater am- the recording was re-referred off-line to linked mastoids
plitude of the second phase of the P600 effect in the (57 and 101). Eye movements and blinks were moni-
sentence experiment. It is worthy of note that these late tored with supra- and infraorbital electrodes and with
differences would depend on the way in which gender electrodes in the external canthi (14, 22 and 125–128).
and number representations affect the reanalysis pro- Interelectrode impedances were kept below 30 k

cesses, rather on the role that these representations play (amplifiers input impedance >200 M
). EEG was fil-
in the early construction of the syntactic structure. tered with an analogue bandpass filter of 0.01–100 Hz
In sum, manipulations of gender and number agree- (50 Hz notch filter) and a digital 35-Hz low-pass filter was
ment resulted in different ERP effects. Anterior negativ- applied before analysis. The signals were sampled con-
ities were found in both experiments and have been tinuously throughout the sessions with a sampling rate
related with the failure of automatic syntactic integra- of 250 Hz.
tion. This effect was accompanied by a more posterior Epochs from the continuous EEG in the interval
N400 effect in the word pair presentation and a P600 between 100 and +900 msec with respect to the onset
effect in the sentence context. While the N400 effect of the target were averaged and analyzed. Baseline
seems to reflect difficulty in the integration of the mor- correction was performed using the average EEG activity
phological features of the words, the P600 effect would in the 100 msec preceding the onset of the target word
be associated to syntactic analysis of the sentence and as a reference signal value. Following baseline correc-
presented two different phases. The late phase was re- tion, trials with artifacts were rejected. Less than 15% of
lated with reanalysis processes and was sensitive to the trials were excluded by this operation and these
the position in which the violation was presented, being were evenly distributed across the different experimen-
greater when the violation was located in the middle of tal conditions. Furthermore, electrodes with a high level
the sentence. The comparison between gender and of artifacts (>10%) were substituted by the average
number agreement processes showed differences in late value of the group of nearest electrodes. Separate ERPs
measures, that is, gender disagreement produced longer were formed for each of the experimental conditions,
latencies of the P3 component in the word pair exper- each of the subjects and each of the electrode sites.
iment and greater amplitudes of the P600 second phase Six regions of interest were computed out of the
in the sentence experiment. These late effects are con- 129 electrodes, each containing the mean of a group
gruent with the proposal that reanalysis processes are of 11 electrodes. The regions were (see electrode num-
costlier after detection of gender violations because of bers in Figure 7): left anterior (13, 20, 21, 25, 28, 29, 30,
the lexical nature of grammatical gender. 34, 35, 36, and 40), left central (31, 32, 37, 38, 41, 42, 43,
46, 47, 48, and 50), left posterior (51, 52, 53, 54, 58, 59,
60, 61, 66, 67, and 72), right anterior (4, 111, 112, 113,
METHODS 116, 117, 118, 119, 122, 123, and 124), right central (81,
88, 94, 99, 102, 103, 104, 105, 106, 109, and 110), and
Participants right posterior (77, 78, 79, 80, 85, 86, 87, 92, 93, 97, and
Twenty-two undergraduates (11 women and 11 men) 98). Additional analyses were carried out for other
participated in Experiment 1, and 24 (18 women and electrode groups, including frontal and occipital regions,
6 men) participated in Experiment 2. All of them were and midline locations. As these revealed similar results
native Spanish speakers, with no history of neurological they are not considered here.
or psychiatric impairment, and with normal or corrected- Different repeated-measures ANOVAs were performed
to-normal vision. Ages ranged from 18 to 26 years introducing these groups of electrodes as the different
(mean = 20.3 years) in Experiment 1, and from 18 to levels of the electrode region factor (anterior, central,

Barber and Carreiras 149

Figure 7. Schematic f lat
representation of the 129
electrode positions from
which EEG activity was
recorded (front of head is at
top). Grouped electrodes in
the analysis (six electrode
regions) and six specific
positions represented in
Figures 2, 4, and 6 are

and posterior) and the hemisphere factor (left and second list, word pairs included a noun followed by an
right), and the experimental variables as additional adjective. The combination of the different gender and
factors. Effects related with electrode region factor or number forms of articles and adjectives resulted in the
hemisphere factor will only be reported when they in- different experimental conditions:
teract with the experimental manipulations. In cases of Article–noun word pairs:
interaction of any experimental factor with the electrode
region or hemisphere factors, data were normalized (a) Agreement, e.g., El piano (the m-s piano m-s).
following the vectorial scaled procedure recommended (b) Gender disagreement, e.g., La piano (the f–s

by McCarthy and Wood (1985). Where appropriate, piano m-s).

critical values were adjusted using the Geisser and (c) Number disagreement, e.g., Los piano (the m-p

Greenhouse (1959) correction for violation of the as- piano m-s).

sumption of sphericity. Noun–adjective word pairs:
Behavioral measures are not reported because speed
was not required of the subjects so reaction times (d) Agreement, e.g., Faro alto (lighthouse m-s high m-s).
showed great variability. (e) Gender disagreement, e.g., faro alta (lighthouse m-s
high f-s).
( f ) Number disagreement, e.g., faro altos (lighthouse m-s
Stimuli high m-p).
Experiment 1 Assignment of word pairs to conditions in each list
Two lists of 120 experimental word pairs each were was counterbalanced across participants. Thus, each pair
generated. In the first list, word pairs were made up of occurred three times across subjects, once in each
a determiner article followed by a noun, whereas in the condition, so that each subject only saw one form of

150 Journal of Cognitive Neuroscience Volume 17, Number 1

each pair during the experiment. In Spanish, it is Procedure
mandatory that articles, nouns, and adjectives agree in
Participants were seated comfortably in a darkened
gender and in number. In the present stimuli, gender
sound-attenuated chamber. Stimuli were displayed in
was always a strictly morphosyntactic feature without
black letters against a gray background. All trials were
semantic significance. All nouns and adjectives were
presented in different pseudorandom order for each
morphologically marked in gender and number, that
participant. Participants had to judge the grammatical
is, they ended with the canonical suffixes in Spanish for
agreement between words of a given pair (Experiment 1)
gender (‘‘–o’’ for masculine and ‘‘–a’’ for feminine) and
or to what extent a sentence was grammatically correct
number (‘‘–s’’ or ‘‘–es’’ for plural).
(Experiment 2). Participants in Experiment 1 had to
In addition, a list of 80 filler trials was introduced.
make the decision after reading the second word, but
Some fillers had opaque gender (e.g., the word ‘‘reloj’’
they were not encouraged to respond quickly. In Exper-
[clock] lacks any explicit morphological mark) and some
iment 2, the response was delayed until the reading of
other fillers were irregular words (e.g., ‘‘mano’’ [hand]
the last word of the sentence. The grammatical judg-
ends with the letter ‘‘–o’’ but is feminine). This type of
ment task was chosen taking into consideration previous
fillers was included to prevent participants using a su-
studies, both with word pairs (Münte & Heinze, 1994)
perficial strategy for solving the task such as, for exam-
and with sentences (Osterhout & Mobley, 1995), which
ple, attending only to the suffixes. All the filler word
compared the effects of agreement violations with or
pairs agreed in gender and number.
without task. In these studies, the same qualitative ef-
fects were found but they were quantitatively greater
Experiment 2 with the grammatical judgment task. A response button
was positioned beneath each thumb. For half of the par-
Two lists of 120 experimental sentences each were
ticipants, the right button was used to signal the ‘‘Yes’’
generated in order to manipulate the agreement at the
response and the left button was assigned to the ‘‘No’’
beginning and in the middle of the sentence. The first
response. For the remaining subjects, the order was re-
list manipulated the agreement relationships at the
versed. Participants were also asked to avoid eye move-
beginning of the sentence using the article–noun word
ments and blinks during the interval when a fixation
pairs list from Experiment 1, while the second list
asterisk was not present.
manipulated the agreement relationships in the middle
The sequence of events in each trial in each experi-
of the sentence using the noun–adjective word pairs list
ment is described as follows:
from Experiment 1:
Manipulation at the beginning of the sentence:
(a) Agreement, e.g., El piano estaba viejo y desafinado Experiment 1
(the m-s piano m-s was old and off-key).
(b) Gender disagreement, e.g., La piano estaba viejo y First, a fixation point (‘‘+’’) appeared in the center of
desafinado (the f–s piano m-s was old and off-key). the screen and remained there for 2700 msec. This fix-
(c) Number disagreement, e.g., Los piano estaba viejo y ation point was followed by a blank screen interval of
desafinado (the m-p piano m-s was old and off-key). 300 msec, then the prime word appeared for 300 msec,
which was followed by a 500-msec blank interval, and
Manipulation in the middle of the sentence: finally the target word appeared and remained there
(d) Agreement, e.g., El faro es alto y luminoso (the m-s up to a maximum of 2000 msec or until the participant’s
lighthouse m-s is high m-s and bright). response. The intertrial interval varied randomly be-
(e) Gender disagreement, e.g., El faro es alta y luminoso tween 1000 and 1500 msec.
(the m-s lighthouse m-s is high f–s and bright).
(f ) Number disagreement, e.g., El faro es altos y lumi-
noso (the m-s lighthouse m-s is high m-p and bright). Experiment 2
Between two and four filler words were included after First, a fixation point (‘‘+’’) appeared in the center of
the target adjective in order to avoid wrap-up effects on the screen and remained there for 2700 msec. This
the target words. As in Experiment 1, assignment of fixation point was followed by a blank screen interval
sentences to conditions in each list was counterbalanced of 300 msec, then the sentence was displayed word by
across participants. Thus, each sentence occurred three word. Each word appeared for 300 msec and was
times across subjects, once in each condition, so that followed by a 300-msec blank interval. Participants were
each subject only saw one form of each sentence during instructed to respond after the last word of the sen-
the experiment. tence. At that moment, a question mark was presented
In addition, a list of 160 well-formed filler sentences and remained there up to a maximum of 2000 msec or
was introduced. Fillers included nouns with opaque or until the participant’s response. The intertrial interval
irregular gender, and adjectives with neuter gender. varied randomly between 1000 and 1500 msec.

Barber and Carreiras 151

Acknowledgments Friederici, A. D., Mecklinger, A., Spencer, K. M., Steinhauer, K.,
& Donchin, E. (2001). Syntactic parsing preferences and
This research was partially supported by fellowship EX2003- their on-line revisions: A spatio-temporal analysis of
0012 (Secretarı́a de Estado de Educación y Universidades of event-related brain potentials. Cognitive Brain Research,
the Spanish Government and the European Social Fund) to the 11, 305–323.
first author, and grant BSO2003-01135 (Spanish Ministry of Garrett, M. F. (1982). Levels of processing in sentence
Science and Technology) to the second author. We thank Lee production. In B. Butterworth (Ed.), Language production
Osterhout and two anonymous reviewers for helpful com- (pp. 170–220). London: Academic Press.
ments, and Margaret Gillon Dowens for her revision of the Geisser, S., & Greenhouse, S. (1959). On methods in the
manuscript. analysis of profiledata. Psychometrika, 24, 95–112.
Reprint requests should be sent to Horacio Barber, PhD, Gunter, T. C., Friederici, A. D., & Schriefers, H. (2000).
University of California, San Diego, Department of Cognitive Syntactic gender and semantic expectancy: ERPs reveal early
Science, 9500 Gilman Drive, La Jolla, CA 92093-0515, or via autonomy and late interaction. Journal of Cognitive
e-mail: Neuroscience, 12, 556–568.
Hagoort, P., & Brown, C. (1999). Gender electrified: ERP
evidence on the syntactic nature of gender processing.
REFERENCES Journal of Psycholinguistic Research, 28, 6, 715–728.
Hagoort, P., & Brown, C. (2000). ERP effects of listening to
Antón-Méndez, I., Nicol, J. L., & Garret, M. F. (2002). The speech compared to reading: The P600/SPS to syntactic
relation between gender and number agreement processing. violations in spoken sentences and rapid serial visual
Syntax, 5, 1–25. presentation. Neuropsychologia, 38, 1531–1549.
Barber, H., & Carreiras, M. (2003). Integrating gender and Hagoort, P., Brown, C., & Groothusen, J. (1993). The syntactic
number information in Spanish word pairs: An ERP study. positive shift (SPS) as an ERP-measure of syntactic
Cortex, 39, 465–482. processing. Language and Cognitive Processes, 8, 439– 483.
Barber, H., Salillas, E., & Carreiras, M. (2004). Gender or Hagoort, P., Brown, C., & Osterhout, L. (1999). The
genders agreement? In M. Carreiras & C. Clifton (Eds.), neurocognition of syntactic processing. In C. Brown &
On-line study of sentence comprehension; eye-tracking, P. Hagoort (Eds.), Neurocognition of language
ERP and beyond. Brighton, UK: Psychology Press. (pp. 273–316). Oxford: Oxford University Press.
Bentin, S., McCarthy, G., & Wood, C. C. (1985). Event-related Harris, J. W. (1991). The exponence of gender in Spanish.
potentials, lexical decision and semantic priming. Linguistic Inquiry, 22, 27–62.
Electroencephalography and Clinical Neurophysiology, Igoa, J. M., Garcı́a-Albea, J. M., & Sánchez-Casas, R. (1999).
60, 343–355. Gender–number dissociation in sentence production in
Bock, K., & Levelt, W. (1994). Language production: Spanish. Rivista di Linguistica, 11, 163–193.
Grammatical encoding. In M. A. Gernsbacher (Ed.), Kaan, E., Harris, A., Gibson, G., & Holcomb, P. J. (2000).
Handbook of psycholinguistics (pp. 945–984). San Diego, The P600 as an index of syntactic integration difficulty.
CA: Academic Press. Language and Cognitive Processes, 15, 159–201.
Colé, P., & Seguı́, J. (1994). Grammatical incongruency and Kluender, R., & Kutas, M. (1993). Bridging the gap:
vocabulary types. Memory and Cognition, 22, 387–394. Evidence from ERPs on the processing of unbounded
Corbett, G. G. (1991). Gender. Cambridge: Cambridge dependencies. Journal of Cognitive Neuroscience, 5,
University Press. 196–214.
Coulson, S., King, J., & Kutas, M. (1998). Expect the Kutas, M., & Federmeier, K. D. (2000). Electrophysiology
unexpected: Event-related brain response to reveals semantic memory use in language comprehension.
morphosyntactic violations. Language and Cognitive Trends in Cognitive Sciences, 4, 463– 470.
Processes, 13, 21–58. Kutas, M., & Hillyard, S. A. (1983). Event-related brain
De Vincenzi, M. (1999). Differences between the morphology potentials to grammatical errors and semantic anomalies.
of gender and number: Evidence from establishing Memory and Cognition, 11, 539–550.
coreferences. Journal of Psycholinguistic Research, 28, Kutas, M., McCarthy, G., & Donchin, E. (1977). Augmenting
5, 537–553. mental chronometry: The P300 as a measure of stimulus
De Vincenzi, M., & Di Domenico, E. (1999). A distinction evaluation time. Science, 197, 792–795.
among phi-features: The role of gender and number in the Lukatela, G., Kostic, A., Todorovic, D., Carello, C., &
retrieval of pronoun antecedents. Rivista di Linguistica, 11, Turvey, M. T. (1987). Type and number of violations and
41–74. the grammatical congruency effect in lexical decision.
Donchin, E. (1979). Event-related brain potentials: A tool in the Psychological Research, 49, 37– 43.
study of human information processing. In H. Begleiter McCarthy, G., & Wood, C. (1985). Scalp distributions of
(Ed.), Evoked potentials and behaviour. New York: Plenum. event-related potentials: An ambiguity associated with
Donchin, E., & Coles, M. (1988). Is the P300 component a analysis of variance models. Electroencephalography and
manifestation of context updating? Behavioral and Brain Clinical Neurophysiology, 62, 203–208.
Sciences, 11, 357–374. Münte, T. F., & Heinze, H. J. (1994). ERP negativities during
Faussart, C., Jakubowicz, C., & Costes, M. (1999). Gender and syntactic processing of written words. In H.-J. Heinze,
number processing in spoken French and Spanish. Rivista di T. F. Münte, & G. R. Mangun (Eds.), Cognitive
Linguistica, 11, 75–101. electrophysiology (pp. 211–238). Boston: Birkäuser.
Frazier, L. (1987). Sentence processing: A tutorial review. Münte, T. F., Matzke, M. & Johannes, S. (1997). Brain activity
In M. Coltheart (Ed.), Attention and performance. associated with syntactic incongruencies in words and
Hillsdale, NJ: Erlbaum. pseudo-words. Journal of Cognitive Neuroscience, 9,
Friederici, A. D. (1995). The time course of syntactic 318–329.
activation during language processing: A model based on Oldfield, R. C. (1971). The assessment and analysis of
neuropsychological and neurophysiological data. Brain handedness: The Edinburgh Inventory. Neuropsychologia,
and Language, 50, 259–281. 9, 97–113.

152 Journal of Cognitive Neuroscience Volume 17, Number 1

Osterhout, L., & Hagoort, P. (1999). A superficial resemblance words: Behavioral and electrophysiological evidence.
does not necessarily mean you are part of the family: Biological Psychology, 48, 183–204.
Counterarguments to Coulson, King and Kutas (1998) in the Ritter, E. (1988). A head-movement approach to construct
P600/SPS–P300 debate. Language and Cognitive Processes, state noun phrases. Linguistics, 26, 909–929.
14, 1–14. Ritter, E. (1993). Where is gender? Linguistic Inquiry, 24,
Osterhout, L., & Holcomb, P. J. (1995). Event related potentials 795–803.
and language comprehension. In M. D. Rugg & M. Coles Rugg, M. D. (1985). The effects of semantic priming and word
(Eds.), Electrophysiology of mind, event related brain repetition on event related potentials. Psychophysiology,
potentials and cognition. Oxford: Oxford University Press. 22, 642–647.
Osterhout, L., & Mobley, L. A. (1995). Event-related potentials Rugg, M. D., & Barrett, S. E. (1987). Event-related potentials
elicited by failure to agree. Journal of Memory and and the interaction between orthographic and phonological
Language, 34, 739–773. information in a rhyme-judgment task. Brain and
Pulvermüller, F. (1999). Words in the brain’s language. Language, 32, 336–361.
Behavioral and Brain Sciences, 22, 253–336. Vigliocco, G., Butterworth, B., & Garrett, M. F. (1996).
Radeau, M., Besson, M., Fonteneau, E., & Luis Castro, S. (1998). Subject–verb in Spanish and English: Differences in the
Semantic, repetition and rime priming between spoken role of conceptual factors. Cognition, 51, 261–298.

Barber and Carreiras 153